A Multi-attribute Controllable Generative Model for Histopathology Image Synthesis

Generative models have been applied in the medical imaging domain for various image recognition and synthesis tasks. However, a more controllable interpretable model is still lacking yet necessary important applications such as assisting training. In this work, we leverage efficient self-attention contrastive learning modules build upon state-of-the-art generative adversarial networks (GANs) to achieve an attribute-aware model, termed AttributeGAN, which can generate high-quality histopathology images based on multi-attribute inputs. comparison existing single-attribute conditional models, our proposed better reflects input attributes enables smoother interpolation among attribute values. We conduct experiments dataset containing stained H&E of urothelial carcinoma demonstrate effectiveness via comprehensive quantitative qualitative comparisons with well different variants model. Code available at https://github.com/karenyyy/MICCAI2021AttributeGAN.